A5102939766- Human Pose and Action Recognition
- Hand Gesture Recognition Systems
- Robot Manipulation and Learning
- Advanced Vision and Imaging
- Stroke Rehabilitation and Recovery
- Context-Aware Activity Recognition Systems
- Anomaly Detection Techniques and Applications
- 3D Shape Modeling and Analysis
- Advanced Neural Network Applications
- Handwritten Text Recognition Techniques
- Video Surveillance and Tracking Methods
- 3D Surveying and Cultural Heritage
- Multimodal Machine Learning Applications
- Digital Radiography and Breast Imaging
- Advanced Optical Sensing Technologies
- Mobile Health and mHealth Applications
- Artificial Intelligence in Healthcare
- User Authentication and Security Systems
- Advanced X-ray and CT Imaging
- Domain Adaptation and Few-Shot Learning
- Image Processing and 3D Reconstruction
Islamic University of Madinah
2018-2023
German Research Centre for Artificial Intelligence
2017-2020
Articulated hand pose and shape estimation is an important problem for vision-based applications such as augmented reality animation.In contrast to the existing methods which optimize only joint positions, we propose a fully supervised deep network learns jointly estimate full 3D mesh representation from single depth image.To this end, CNN architecture employed parametric representations i.e. pose, bone scales complex parameters. Then, novel layer, embedded inside our framework, produces...
Realistic reconstruction of two hands interacting with objects is a new and challenging problem that essential for building personalized Virtual Augmented Reality environments. Graph Convolutional networks (GCNs) allow the preservation topologies poses shapes by modeling them as graph. In this work, we propose THOR-Net which combines power GCNs, Transformer, self-supervision to realistically reconstruct an object from single RGB image. Our network comprises stages; namely features extraction...
In-air signature verification is vital for biometric user identification in contact-less mode. The state-of-the-art methods use heuristics acquisition, and provide insufficient data to train neural networks the verification. In this article, we present a novel method end-to-end deep learning based in-air using depth sensor. regard, propose new medium-scale dataset which created an accurate convolutional network (CNN) 3D hand pose estimation algorithm. proposed offers total of 1800 signatures...
In-air signature is a new modality which essential for user authentication and access control in noncontact mode has been actively studied recent years. However, it treated as conventional online signature, essentially 2D spatial representation. Notably, this bears lot more potential due to an important hidden depth feature. Existing methods in-air verification neither capture unique feature explicitly nor fully explore its verification. Moreover, these are based on heuristic approaches...
3D hand shape and pose estimation from a single depth map is new challenging computer vision problem with many applications. Existing methods addressing it directly regress meshes via 2D convolutional neural networks, which leads to artifacts due perspective distortions in the images. To address limitations of existing methods, we develop HandVoxNet++, i.e., voxel-based deep network graph convolutions trained fully supervised manner. The input our voxelized-depth-map-based on truncated...
Hand shape and pose recovery is essential for many computer vision applications such as animation of a personalized hand mesh in virtual environment. Although there are estimation methods, only few deep learning based algorithms target 3D from single RGB or depth image. Jointly estimating very challenging because none the existing real benchmarks provides ground truth shape. For this reason, we propose novel weakly-supervised approach (named WHSP-Net) image by shapes unlabeled data labeled...
Articulated hand pose estimation is a challenging task for human-computer interaction. The state-of-the-art algorithms work only with one or few subjects which they have been calibrated trained. Particularly, the hybrid methods based on learning followed by model fitting deep do not explicitly consider varying shapes and sizes. In this work, we introduce novel algorithm estimating 3D as well bone-lengths of skeleton at same time, from single depth image. proposed CNN architecture learns...
The estimation of human hand pose has become the basis for many vital applications where user depends mainly on as a system input. Virtual reality (VR) headset, shadow dexterous and in-air signature verification are few examples that require to track movements in real-time. state-of-the-art 3D methods based Convolutional Neural Network (CNN). These implemented Graphics Processing Units (GPUs) due their extensive computational requirements. However, GPUs not suitable practical application...
Recovery of articulated 3D structure from 2D observations is a challenging computer vision problem with many applications. Current learning-based approaches achieve state-of-the-art accuracy on public benchmarks but are restricted to specific types objects and motions covered by the training datasets. Model-based do not rely data show lower these In this paper, we introduce model-based method called Structure Articulated Motion (SfAM), which can recover multiple object motion without...
Estimating the hand-object meshes and poses is a challenging computer vision problem with many practical applications. In this paper, we introduce simple yet efficient reconstruction algorithm. To end, exploit fact that both are graphs-based representations of different levels details. This allows taking advantage powerful Graph Convolution networks (GCNs) to build coarse-to-fine Graph-based Thus, start by estimating coarse graph represents 2D poses. Then, more details (e.g. third dimension...
Skeleton-based human action recognition is a challenging yet important technique because of its wide range applications in many fields, including patient monitoring, security surveillance, and observing human-machine interactions. Many algorithms that attempt to distinguish between types activities have been proposed. However, most practical require highly accurate detection specific activities. In this study, novel spatiotemporal graph autoencoder network for skeleton-based Furthermore, an...
The task of human action recognition (HAR) based on skeleton data is a challenging yet crucial technique owing to its wide-ranging applications in numerous domains, including patient monitoring, security surveillance, and observation human-machine interactions. While algorithms have been proposed an attempt distinguish between myriad activities, most practical necessitate highly accurate detection specific activity types. This study proposes novel spatiotemporal graph autoencoder network for...
3D reconstruction of hand-object manipulations is important for emulating human actions. Most methods dealing with challenging object manipulation scenarios, focus on hands in isolation, ignoring physical and kinematic constraints due to contact. Some approaches produce more realistic results by jointly reconstructing interactions. However, they coarse pose estimation or rely upon known hand shapes. We propose the first approach shape from a single depth map. Unlike previous work, our...
Articulated hand pose and shape estimation is an important problem for vision-based applications such as augmented reality animation. In contrast to the existing methods which optimize only joint positions, we propose a fully supervised deep network learns jointly estimate full 3D mesh representation from single depth image. To this end, CNN architecture employed parametric representations i.e. pose, bone scales complex parameters. Then, novel layer, embedded inside our framework, produces...
3D hand shape and pose estimation from a single depth map is new challenging computer vision problem with many applications. Existing methods addressing it directly regress meshes via 2D convolutional neural networks, which leads to artefacts due perspective distortions in the images. To address limitations of existing methods, we develop HandVoxNet++, i.e., voxel-based deep network graph convolutions trained fully supervised manner. The input our voxelized-depth-map-based on truncated...
3D reconstruction of hand-object manipulations is important for emulating human actions. Most methods dealing with challenging object manipulation scenarios, focus on hands in isolation, ignoring physical and kinematic constraints due to contact. Some approaches produce more realistic results by jointly reconstructing interactions. However, they coarse pose estimation or rely upon known hand shapes. We propose the first approach shape from a single depth map. Unlike previous work, our...
Human motion recognition is extremely important for many practical applications in several disciplines, such as surveillance, medicine, sports, gait analysis, and computer graphics.Graph convolutional networks (GCNs) enhance the accuracy performance of skeleton-based action recognition.However, this approach has difficulties modeling long-term temporal dependencies.In Addition, fixed topology skeleton graph not sufficiently robust to extract features motions.Although transformers that rely...